# Gensyn > Independent Gensyn review. ML training protocol on a custom Ethereum L2 rollup. $AI token TGE 29 April 2026, Delphi market live, training mainnet pending. - Scores: Freedom 5.1/10 (grade D), Returns 3.8/10 (grade F) - Category: compute · Token: AI - Published: 2026-03-02 · Updated: 2026-09-29 - Canonical: https://ownyourmind.ai/projects/gensyn/ - Data (scores + integrity verdicts, JSON): https://ownyourmind.ai/api/projects/gensyn.json - Cite as: Own Your Mind (ownyourmind.ai), licence CC BY 4.0. Link the canonical URL. Charts and interactive widgets are omitted here; their data is in the JSON above. --- ## What it does Gensyn is a compute protocol designed specifically for machine learning training. ML training only, with a verification layer to prove the work was done correctly. General compute, rendering and storage are out of scope. The architecture has four layers. At the bottom, the SAPO algorithm ensures ML computations produce identical results across different hardware, whether that is an NVIDIA A100, an RTX 4090, or an Apple M-series chip. Above that, the Verde verification system uses Reproducible Operators (RepOps) for bitwise-identical ML outputs, with a two-level bisection game for dispute resolution. The peer-to-peer layer handles gradient sharing without centralised coordinators (NoLoCo for gossip-based gradient averaging, CheckFree for fault-tolerant recovery, SkipPipe for efficient gradient sharing). The coordination layer is a custom Ethereum L2 rollup that handles participant registration, task submission, proof verification, and payment settlement. Four roles operate the network: Submitters (who create training tasks), Solvers (who provide compute), Verifiers (who check the work), and Whistleblowers (who challenge incorrect verifications). The actual ML training happens off-chain. The proofs, payments, and dispute resolution happen on-chain. The project was co-founded in May 2020 by Ben Fielding (PhD in AI/ML from Northumbria, research in evolutionary neural architecture search) and Harry Grieve (MA Economics from Aberdeen, Master of Public Affairs from Brown, former Director of Data Research at Cytora). They met through Entrepreneur First's London cohort. The company, Gensyn Limited, is registered at Companies House (#12601008) in London. Team size is 44 with 20-plus open roles. Funding totals $66.74 million across four rounds: pre-seed ($1.1M, January 2021), seed ($6.5M, March 2022, led by Eden Block), Series A ($43M, June 2023, led by a16z crypto), and a public token sale ($16.14M, December 15-20 2025 via English auction on Sonar, 7,412 participants, clearing at $0.0473/token and $473M FDV). Other backers include CoinFund, Protocol Labs, Canonical Crypto, Maven 11 Capital, and Galaxy. Current status: the Gensyn L2 is live, with **Delphi (AI-settled prediction markets) launched on 22 April 2026** as the first product on the network. The $AI token went through TGE on **29 April 2026**, hitting an all-time high of $0.1034 the same day before falling roughly 68% in the following three days. Trading is now active on Bitget, KuCoin, Kraken, MEXC, Uniswap V3/V4 and Binance Alpha, with a Binance perpetual added shortly after. Reported daily volume has since fallen roughly seven-eighths from the TGE window. The training network itself, the original product the protocol was designed around, is still pre-mainnet. Testnet (Phase 0, launched March 2025) continues training 0.5B-1.5B Qwen variants. ## Value proposition The pitch is simple. ML training is expensive, GPU supply is constrained, and most of the world's compute sits idle. Gensyn connects underutilised hardware into a unified training network with cryptographic verification that the work was done correctly. What makes Gensyn stand apart from the dozen other "decentralised GPU" projects is the research depth. Its papers on arXiv cover: - Verde (verification) - NoLoCo (gradient averaging) - CheckFree (fault tolerance) - SkipPipe (gradient sharing) - SAPO (collective RL post-training) - Reinforcement learning swarms and diverse network ensembles - Open-1B (auditable training, September 2026) The Verde system, which enables bitwise-reproducible ML across heterogeneous hardware, is a novel contribution to the field. a16z's investment thesis called them "nobody better" at combining AI and crypto. The testnet drew wide engagement while it ran. Gensyn reports 165,000-plus users, two million models trained, and 90 million transactions. RL Swarm let anyone with a GPU or 32GB of RAM run a node and join collaborative reinforcement learning training. Since 5 January 2026 Gensyn's docs have read "There are no official swarms running right now", and the RL Swarm page is now titled "RL Swarm (Paused)". Delphi is the one live product: a prediction market where users stake on which AI models will perform best on benchmarks. CodeAssist, which trained coding models locally on your own edits, and BlockAssist, its Minecraft counterpart, stopped tracking new testnet participation on 2 March 2026 "as we shift focus to Mainnet". Gensyn's docs now describe both as sunset and no longer maintained. The counter-narrative is equally clear. The training network itself is still on testnet. The models being trained are 0.5B to 1.5B parameter Qwen variants, not frontier-scale workloads. The litepaper projected V100-equivalent pricing at $0.40/hour (80% cheaper than AWS), but no compute pricing exists yet. Delphi is the first product to ship to the L2, but it is a prediction market, not a training service. Competitors like [Akash](https://ownyourmind.ai/projects/akash/) and [io.net](https://ownyourmind.ai/projects/io-net/) have production networks handling paying workloads today. Gensyn is still proving that its verification system works at scale. The ML-specific focus is both a strength and a limitation. General compute networks can serve ML workloads plus everything else. Gensyn is ML-only. If the verification layer and distributed training primitives work at production scale, the specialisation is a moat. If not, it's a constraint. ## Tokenomics $AI is an ERC-20 token on the Gensyn L2. Total supply: 10 billion. The token sale closed in December 2025, with tokens distributed to buyers in early 2026. 1.31 billion tokens (13.05%) are now in circulation as of June 2026. Distribution: - **Community Treasury:** 40.4% (20% unlocked at TGE, remainder linear over 36 months). Funds ecosystem development, grants, and programmes. Controlled by the Gensyn Foundation. - **Investors:** 29.6% (12-month cliff, then 24-month linear unlock). Pre-seed, seed, and Series A investors. Cannot stake during lockup. - **Team:** 25.0% (12-month cliff, then 24-month linear unlock). Cannot stake during lockup. - **Community Sale:** 3.0% (immediate unlock for non-US buyers; 12-month lockup for US buyers and those opting for 10% bonus multiplier). 300 million tokens sold via English auction at the clearing price. - **Testnet Rewards:** 2.0% (distributed based on testnet participation level and sale bid amount). The insider allocation is 54.6% (investors plus team). That is high. The 12-month cliff and 24-month linear unlock provide meaningful protection, and locked tokens cannot be staked. But once the cliff expires (approximately April 2027), 24 months of continuous insider selling pressure begins. The community's direct access is limited: 3% public sale plus 2% testnet rewards equals 5% of total supply. The remaining 40.4% community treasury sounds generous but is controlled by the Foundation, which is initially team-aligned. Token utility spans five functions on paper: compute payments, verification staking, Delphi prediction market stakes, governance voting, and a programmatic buy-and-burn mechanism from transaction fees. Reading the deployed contracts moves one of those off the list. Delphi stakes and settles in bridged USDC, not in $AI: the same USDC address answers `TOKEN()` on both the trading gateway and the market factory. So $AI is not what users put at risk on a Delphi market, and the compute-payment leg still waits on a training-network mainnet that has not shipped. What is live is the burn, and it is more than a design. Delphi's fees accrue in USDC to a BuybackVault, which swaps them for $AI on the network's own Uniswap router and destroys 70% of the proceeds, a figure the contract states directly as `burnBps` 7000. One per cent goes to whoever triggers the round and the rest to treasury. It has run 17 times since May 2026, roughly weekly. (figures as of 2026-09-21) The scale is the part to hold on to. Those rounds have converted $4,499 of lifetime fees and burned 128,655 $AI, which is 0.0013% of the cap. The burn total matched the gap between the ten-billion cap and live supply when we reconciled it in August 2026. A revenue-funded burn that works, moving four thousand dollars, is a mechanism attached to almost no economy. (figures as of 2026-09-21) The public sale used a transparent English auction on Sonar. The token began trading on 29 April 2026 and now sits on Bitget, KuCoin, Kraken, MEXC, Uniswap V3/V4 and Binance Alpha, with a 20x Binance perpetual added shortly after. Reported daily volume has fallen roughly seven-eighths from the TGE window, per CoinGecko in September 2026. Price action has been brutal: ATH of $0.1034 on TGE day, down roughly 68% within three days. ## How to participate **Join a community RL Swarm.** Gensyn has run no official swarm since 5 January 2026, so this is now a community exercise rather than Gensyn's testnet. The code is still open and permissionless. Install Docker, Python 3.10-plus, and Node.js 22.x. GPU option: NVIDIA RTX 3090, 4090, 5090, A100, or H100 with 12GB-plus VRAM and CUDA 12.4-plus. CPU option: ARM64 or x86 with 32GB-plus RAM (slower but functional). Nodes train small Qwen2.5 models (0.5B to 1.5B parameters) in a distributed reinforcement learning swarm. **Predict on Delphi.** The prediction market lets you stake on which AI models will outperform on benchmarks. Markets stake and settle in bridged USDC rather than $AI. Stake on expected winners, earn rewards when markets settle. **Build.** Contribute to open-source repos (rl-swarm, rl-swarm-contracts). rl-swarm is MIT licensed. codeassist (MIT) and blockassist (no licence file) are still public but Gensyn no longer maintains them. Deploy custom swarms. Publish models to HuggingFace. The documentation is solid, covering protocol design, testnet setup, and product guides. The arXiv papers provide deep technical context. **Trade or hold AI.** The token began trading on 29 April 2026 across Bitget, KuCoin, Kraken, MEXC, Uniswap V3/V4 and Binance Alpha. A Binance perpetual with 20x leverage was added shortly after. Reported daily volume has fallen roughly seven-eighths since launch week. Note the post-TGE volatility: AI hit an all-time high of $0.1034 on launch day before falling roughly 68% in three days. Compute-fee revenue and the buy-and-burn at scale still depend on the training-network mainnet that has not yet shipped. ## Honest assessment ### What works The research is substantive. An arXiv record of this depth from a team that includes academic researchers and cryptographers is exceptional for a crypto project. Verde verification (bitwise-reproducible ML across heterogeneous hardware) is a substantive technical contribution that no competitor has replicated. The founding team is credible: Fielding's PhD work directly feeds into the protocol design, and Grieve brings quantitative finance expertise. Both are doxxed, verified through Companies House, and have been building since May 2020. The testnet engagement was solid while it ran. 165,000-plus users and two million models trained (per Gensyn's reported metrics) suggests grassroots interest, even accounting for airdrop farming. The RL Swarm code still runs distributed reinforcement learning on consumer hardware, though Gensyn itself has run no official swarm since January 2026. Open-source credentials are strong. MIT licence on the flagship repos. Over 1,600 stars and 600 forks on rl-swarm as of September 2026. Six models published on HuggingFace. The research-to-product pipeline is visible: academic papers become protocol components become testnet features. Open-1B, released on 15 September 2026, applies the same reproducibility idea to a whole pretraining run. The preprint (arXiv 2609.17380, nine authors including Fielding and Grieve) fixes the order of three sources of training nondeterminism: GPU kernel reductions, data-batch ordering across a data-parallel cluster, and inter- and intra-node collective communication. With that order fixed, any single step of a distributed run can be replayed bitwise on one piece of commodity hardware and checked against the published trajectory, while many independent auditors each certifying individual steps cover the whole run. Gensyn released the model with its full pretraining dataset, every intermediate checkpoint, the training codebase and the audit tooling needed to replay any step. The paper calls this a new tier of model transparency. It's an arXiv preprint and hasn't been through peer review. The test for the Open Source score is whether anyone outside Gensyn picks up the audit tooling, and Open-1B has no bearing on the training-network mainnet that Returns waits on. ### What does not work The training-network mainnet, the project's whole reason for existing, is still not live. No security audits despite $66.74 million raised over nearly six years. Delphi is shipped and the token is trading, but a prediction market is not the compute product investors paid for. The testnet that stood in for it has been paused since January 2026, and CodeAssist and BlockAssist were sunset in March. The first three days of trading saw a 68% drawdown from the TGE-day peak, which is normal for tokens with thin float, but it makes price discovery noisy. At some point, "still on testnet for the actual product" stops being patience and starts being a red flag. The token distribution is insider-heavy. 54.6% to team and investors is above average for the space. The 40.4% community treasury is controlled by the Foundation, whose council is "initially mapped to core team members." In practice, the team controls the vast majority of tokens at launch. Governance doesn't exist. No DAO, no on-chain voting, no governance forum. The Foundation describes "progressive decentralisation" as a plan, not a feature. Two directors (Fielding and Grieve) make all decisions. For a project raising from the public via token sales, the absence of any community governance mechanism is a notable gap. Testnet trains small models. Qwen 0.5B to 1.5B parameters is useful for reinforcement learning research but nowhere near the frontier-scale training (70B-plus parameters) that would justify the "network for machine intelligence" branding. Scaling from testnet to production is an unproven leap. ### The risk Execution risk dominates. Transitioning from a testnet training small RL models to a mainnet handling live economic value and production-grade ML workloads is a massive engineering challenge. The Verde verification system adds computational overhead to training; at production scale, this overhead needs to be manageable. Competition isn't waiting. [Akash](https://ownyourmind.ai/projects/akash/) has had a production compute marketplace since 2020. [io.net](https://ownyourmind.ai/projects/io-net/) is delivering millions of compute hours. Even within the ML-specific niche, Bittensor subnets compete for the same workloads. Gensyn's verification moat only matters if they ship mainnet before the market decides the problem is solved. Regulatory risk is material. A UK-registered company conducting a public token sale with US buyer restrictions faces securities classification risk in multiple jurisdictions. The 12-month US buyer lockup suggests the team is aware of this risk. The public sale FDV prices perfection. Investors paid for a fully functioning mainnet, broad adoption, and sustained revenue. None of these exist yet. If mainnet slips or adoption is slow, the valuation repricing will be painful. No security audits have been published for the smart contracts or protocol, despite $66.74 million in funding and a testnet Gensyn says has processed 90 million transactions. ### My position I don't hold $AI tokens. Trading is now live across multiple exchanges, so taking exposure is straightforward, but I want to see what the training-network mainnet actually delivers before forming a price view. The research depth and technical differentiation stand out, and Delphi shipping ahead of compute is at least a working product on the L2. For now, this is a credible research bet with significant execution, competition, and post-TGE volatility risk. With the official swarms paused since January, Delphi is the only hands-on product Gensyn itself runs; the token offers a position that prices in a future mainnet I haven't yet seen. ## Freedom Score: 51/100 Gensyn scores 51/100 (D grade). Full methodology at [Freedom Score Methodology](https://ownyourmind.ai/resources/freedom-score-methodology/). **Infrastructure decentralisation (10/20):** The RL Swarm software is permissionless. Anyone with qualifying hardware (GPU or 32GB-plus RAM CPU) can run a node, but Gensyn has run no official swarm since 5 January 2026, so today that layer carries only community swarms. The network supports heterogeneous hardware from gaming GPUs to data centre A100/H100s. The Verde verification system enables trustless work verification without centralised intermediaries. However, the coordination layer runs on a custom Ethereum rollup whose sequencer and operator details are not disclosed. Testnet infrastructure relies on Alchemy for authentication. The Gensyn L2 is now live with the Delphi prediction market and a tradeable token, but the training-network mainnet itself is not, so live economic security is partial. **Governance decentralisation (5/20):** Governance is fully centralised. Gensyn Limited (Companies House #12601008) controls all protocol decisions. Two directors (Fielding and Grieve) are the sole decision-makers. Post-TGE plans include an elected council "initially mapped to core team members" with on-chain proposals. This is a stated intention, not a delivered feature. No DAO, no community voting, no governance forum. **Token distribution fairness (6/15):** Insider allocation is 54.6% (29.6% investors plus 25% team). Vesting mitigates concentration risk: 12-month cliff plus 24-month linear unlock, and locked tokens cannot be staked. The community's direct access is 5% (3% public sale plus 2% testnet rewards). The 40.4% community treasury is controlled by the Foundation, which is initially team-aligned. The public sale used a transparent English auction. Not a fair launch. **Censorship resistance (8/15):** The protocol design is permissionless. Any device can contribute compute. Verde verification removes the need to trust individual providers. The Ethereum rollup inherits some censorship resistance from L1 settlement. However, the rollup sequencer is undocumented, creating a potential censorship vector. The $AI token has no pause function, but its logic can be upgraded through a timelocked admin. No censorship incidents on testnet. **Data sovereignty (10/15):** The architecture provides meaningful user control. CodeAssist trained models locally on user machines until it was sunset in March 2026. RL Swarm nodes train local models with results published to user-owned HuggingFace accounts. The litepaper describes optional functional encryption for training on encrypted data. The peer-to-peer architecture avoids centralised data aggregation. Only proofs are submitted on-chain, not raw training data or model weights. **Open source and transparency (12/15):** The flagship repositories rl-swarm (over 1,600 stars) and codeassist (nearly 700 stars) are MIT licensed, while rl-swarm-contracts carries MIT headers in its source without a licence file, blockassist has no licence, and about half of Gensyn's own repositories carry no licence file. Academic papers on arXiv with full methodology, now including the Open-1B auditable training run. Six models published on HuggingFace. Smart contract code is publicly auditable. The research-to-product pipeline is transparent. Deductions: the unlicensed repositories, no public rollup code, no published security audits, and an outdated litepaper. ### Path to improvement Three changes would materially increase Gensyn's score: 1. **Launch mainnet with transparent metrics.** Official testnet swarms have been paused since January 2026. Mainnet launch with a public, real-time dashboard showing active nodes, compute hours delivered, revenue, and verification metrics would be the single biggest credibility signal. Self-reported testnet figures from press releases are not enough. 2. **Implement governance now that TGE has happened.** Token holders who spent real money on $AI tokens should have governance rights. An elected council "initially mapped to core team members" isn't decentralisation. Deploying on-chain governance with community-elected council members, proposal submission, and transparent voting would demonstrate actual commitment to decentralisation. The token sale closed in December 2025; this is overdue. 3. **Publish security audits.** $66.74 million in funding, 90 million testnet transactions by Gensyn's count, and a public token sale, yet no published audit from a recognised firm. Before mainnet handles live economic value, smart contract and protocol audits are essential. This is an easy gap to close with available resources. ## Returns Score: 38/100 AI scores 38/100 (F grade). Full methodology at [Returns Score Methodology](https://ownyourmind.ai/resources/returns-score-methodology/). **Token utility (9/20):** The $AI token has been distributed to sale participants but none of the intended utility is live. Planned use cases span compute payments, verification staking, governance voting, and a programmatic buy-and-burn mechanism. Delphi market stakes were on that list until August 2026, when reading the deployed contracts showed Delphi stakes and settles in bridged USDC rather than $AI, so the one shipped product does not use the token. Scoring generously for design intent, but design intent is not utility, and the September 2026 review took the score down a point for it. **Value accrual (10/20):** The buy-and-burn from transaction fees is the primary value accrual path, and as of August 2026 it is verified rather than assumed. Delphi's USDC fees route to a BuybackVault that swaps for $AI and burns 70%, stated on the contract as `burnBps` 7000, and it has executed 17 times. The burns sum to 128,655 $AI and matched the cap-minus-supply gap at the August 2026 reconciliation. That clears the Tier 1 condition this dimension was held on, which took the score from 6 to 10 at the September 2026 review. The flows behind it are still only four figures of dollars, so 10 is the floor of the band that describes an implemented mechanism rather than an aspirational one. (figures as of 2026-09-21) **Supply dynamics (6/20):** Ten billion tokens with 54.6% allocated to insiders (29.6% investors plus 25% team). The 12-month cliff and 24-month linear unlock provide meaningful short-term protection, and locked tokens cannot be staked, which prevents insiders from compounding their position during the lockup. But once the cliff expires around April 2027, two years of continuous insider selling pressure begins. The community's direct access is just 5% of total supply. The 40.4% community treasury sounds generous until you realise the Foundation, which is initially team-aligned, controls it. **Revenue sustainability (3/25):** Compute revenue is still zero and the training-network mainnet has not shipped. Delphi, live since 22 April 2026, is the first product on the Gensyn L2 generating protocol income, and that income is now measured rather than described: $4,499 of fees converted since May 2026. Its trading fee is set per market and bounded between 0.5% and 5% by the gateway, so the flat "roughly 2%" this review used to quote was never a protocol constant. The litepaper projected V100-equivalent pricing at $0.40 per hour, which would be compelling if it existed. Verifiability is no longer the blocker; scale is. (figures as of 2026-09-21) **Liquidity and access (10/15):** TGE happened on 29 April 2026. The token now trades on Bitget, KuCoin, Kraken, MEXC, Uniswap V3/V4 and Binance Alpha, with a Binance perpetual offering 20x leverage added shortly after. Daily volume has fallen roughly seven-eighths from the TGE window and now turns over around a tenth of market cap, still comfortably enough for retail-scale entry and exit. Coinbase and Binance spot are still missing; the listing surface is strong but not yet tier-one complete. Price action has been ugly (ATH $0.1034 on TGE day, down ~68% within three days), which doesn't change access scoring but signals thin float. ### Path to improvement Three changes would materially increase Gensyn's returns score: 1. **Launch mainnet and generate revenue.** The entire returns case hinges on the buy-and-burn mechanism working with actual transaction fees. Until compute workloads generate revenue on a live network, every tokenomics projection is theoretical. Revenue is the single metric that would transform this score. 2. **Reduce the insider allocation or increase community access.** 54.6% insider allocation with only 5% direct community access is poor by any standard. Increasing the testnet rewards allocation, adding a broader public sale tranche, or committing to community-directed treasury governance would improve the supply dynamics score materially. 3. **Demonstrate pricing competitiveness at scale.** The $0.40/hour V100-equivalent claim needs real-world validation. Publishing transparent pricing benchmarks against AWS, Lambda Labs, and competing decentralised compute networks once mainnet launches would establish whether the economic model actually delivers value to compute buyers. ## Score change log | Date | Score | Change | Reason | | - | - | - | - | | 2026-09-29 | Data | N/A | Correction: RL Swarm official swarms paused since 5 Jan 2026, CodeAssist and BlockAssist sunset 2 Mar 2026 (Gensyn docs and repos); the review had them as live. Infrastructure 10/20 and Data Sovereignty 10/15 held. | | 2026-09-25 | Score | Freedom 52 → 51 (D held) | Open Source 13 → 12: blockassist and rl-swarm-contracts carry no licence file (the rationale called them MIT), about half of Gensyn's own repositories are unlicensed, and the rollup code isn't public. | | 2026-09-24 | Data | N/A | Buyback figures moved off the stale 14-round totals, the three early-May volume figures and the exact-reconciliation claim withdrawn, blockassist's MIT label corrected (no licence), Delphi rounds replaced by USDC settlement, pause claim corrected. Scores unchanged. | | 2026-09-24 | Data | N/A | Added Open-1B (arXiv 2609.17380, a preprint, 15 Sep 2026) and its audit tooling. Dropped the "seven arXiv papers" count, which no longer matches Gensyn's arXiv output. Scores unchanged. | | 2026-09-23 | Data | N/A | Integrity records re-cut: 2 claims now quote Gensyn's own pages verbatim and 7 are our own measurements. Scores unchanged. | | 2026-09-06 | Data | N/A | WS3b distribution rail: on-chain top-10 verdict added on Gensyn's own L2, raw 49.89% and infra-excluded 46.01% across 3,918 holders, with one unlabelled address at 33.97%. Nine of the top ten are unattributed, so the measured 0% insider share records missing labels rather than absent insiders. Token Distribution Fairness 6/15 held. | | 2026-09-06 | Data | N/A | WS3b liquidity rail: the ~$26M daily volume figure (dated early May 2026) has fallen about seven-eighths; restated as turnover near 8% on both surfaces. All named venues confirm across 22 venues. The Binance venue row in the ticker set was NOT treated as evidence of a main-spot listing, since more than one project uses the AI ticker. Liquidity & Access 10/15 held. | | 2026-09-05 | Returns | Token Utility 10 → 9, Value Accrual 6 → 10, Revenue Sustainability 2 → 3 (Returns 34 → 38, F held) | Executes the decision of 2026-08-17 on the on-chain Delphi data. Utility falls because Delphi settles in bridged USDC, not $AI; accrual rises because the buy-and-burn is deployed and running rather than aspirational. | | 2026-08-17 | Data | N/A | Delphi's fee and burn flows read off the Gensyn L2 (chainId 685689, keyless Blockscout). Buy-and-burn is deployed and running: 14 rounds since May 2026 converted ~$4.0K of USDC fees and burned ~111.8K $AI, reconciling exactly with cap-minus-supply. Discharges the "not independently verified" hedge. Flagged for September. | | 2026-08-17 | Data | N/A | Correction: Delphi stakes and settles in bridged USDC, not $AI (`TOKEN()` on both gateway and factory). The review listed "Delphi market stakes" as live $AI utility, which was wrong. Trading fee is per-market, bounded 0.5-5%, not a flat 2%. Token Utility flagged DOWN for September. | | 2026-08-13 | Data | N/A | Supply verdict re-pointed to Gensyn's own $AI token page. Their MiCA whitepaper explicitly reserves future emission adjustments to governance, so the 10B total should not be read as a no-inflation guarantee; noted in the verdict. | | 2026-08-12 | Data | N/A | WS3 on-chain reconciliation (1->4 verdicts). First read of the CANONICAL supply on the Gensyn L2 (10B fixed at init, no mint path, burn-only; match), unblocked by the new coldstart gensyn chain mapping. Token-admin: permissionless transfers, upgrade behind a TimelockController (established). Open-source confirmed (gensyn-ai 37 MIT repos; match). Scores unchanged. | | 2026-06-02 | Data | N/A | Circulating supply corrected from 300M (3%, TGE-day figure) to 1.31B (13.05% of 10B) per CoinGecko/Tokenomist post-TGE unlocks (TokenFactsStrip + body). Score unchanged; pencilled Returns upgrades remain deferred pending Tier-1 on-chain Delphi fee/burn data and the training-network mainnet. June monthly review. | | 2026-05-02 | Returns | 28 → 34 | TGE on 29 April 2026 with multi-exchange trading (Bitget, KuCoin, Kraken, MEXC, Uniswap, Binance Alpha + perpetual). Liquidity & Access 4→10. Other Returns dimensions held pending Tier 1 on-chain Delphi fee/burn data. | | 2026-04-06 | Data | N/A | Updated token sale details (7,412 participants, $0.0473/token, claims opened Feb 2026). Mainnet status unclear despite Nov 2025 "3-4 weeks" claim. Token still not trading. | | 2026-03-12 | Returns | 24 → 28 | Token launched, exchange listings secured. Liquidity & Access 0→4. | | 2025-03-06 | Both | N/A | Initial publish. Freedom 45/100, Returns 24/100. |